4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_internal.h
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
48 #ifndef _MDT_INTERNAL_H
49 #define _MDT_INTERNAL_H
51 #if defined(__KERNEL__)
53 #include <obd_class.h>
54 #include <lustre_net.h>
55 #include <lustre/lustre_idl.h>
56 #include <obd_class.h>
57 #include <lustre_disk.h>
58 #include <lu_target.h>
59 #include <md_object.h>
60 #include <lustre_fid.h>
61 #include <lustre_fld.h>
62 #include <lustre_mdt.h>
63 #include <lustre_req_layout.h>
64 #include <lustre_sec.h>
65 #include <lustre_idmap.h>
66 #include <lustre_eacl.h>
67 #include <lustre_quota.h>
69 /* check if request's xid is equal to last one or not*/
70 static inline int req_xid_is_last(struct ptlrpc_request *req)
72 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
73 return (req->rq_xid == lcd->lcd_last_xid ||
74 req->rq_xid == lcd->lcd_last_close_xid);
79 /* file data for open files on MDS */
80 struct mdt_file_data {
81 struct portals_handle mfd_handle; /* must be first */
82 __u64 mfd_mode; /* open mode provided by client */
83 cfs_list_t mfd_list; /* protected by med_open_lock */
84 __u64 mfd_xid; /* xid of the open request */
85 struct lustre_handle mfd_old_handle; /* old handle in replay case */
86 struct mdt_object *mfd_object; /* point to opened object */
89 #define CDT_NONBLOCKING_RESTORE 0x0000000000000001ULL
90 #define CDT_NORETRY_ACTION 0x0000000000000002ULL
91 #define CDT_POLICY_MASK CDT_NONBLOCKING_RESTORE | \
94 /* when adding a new policy, do not forget to update
95 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
97 #define CDT_DEFAULT_POLICY 0x0000000000000000ULL
99 enum cdt_states { CDT_STOPPED = 0,
105 /* when multiple lock are needed, the lock order is
113 struct ptlrpc_thread *cdt_thread; /**< coordinator thread */
114 struct lu_env cdt_env; /**< coordinator lustre
116 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
117 __u64 cdt_policy; /**< flags to defined
119 enum cdt_states cdt_state; /**< state */
120 cfs_atomic_t cdt_compound_id; /**< compound id counter */
121 __u64 cdt_last_cookie; /**< last cookie allocated */
122 struct semaphore cdt_counter_lock; /**< protect request
124 struct semaphore cdt_llog_lock; /**< protect llog access */
125 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
126 struct rw_semaphore cdt_request_lock; /**< protect request list */
127 struct semaphore cdt_restore_lock; /**< protect restore list */
128 cfs_time_t cdt_loop_period; /**< llog scan period */
129 cfs_time_t cdt_delay; /**< request grace delay */
130 cfs_time_t cdt_timeout; /**< request timeout */
131 __u64 cdt_max_request; /**< max count of started
133 __u64 cdt_request_count; /**< current count of
134 * started requests */
135 cfs_list_t cdt_requests; /**< list of started
137 cfs_list_t cdt_agents; /**< list of register
139 cfs_list_t cdt_restore_hdl; /**< list of restore lock
143 /* mdt state flag bits */
144 #define MDT_FL_CFGLOG 0
145 #define MDT_FL_SYNCED 1
149 struct lu_device mdt_lu_dev;
150 struct seq_server_site mdt_seq_site;
151 /* DLM name-space for meta-data locks maintained by this server */
152 struct ldlm_namespace *mdt_namespace;
153 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
154 struct ptlrpc_client *mdt_ldlm_client;
155 /* underlying device */
156 struct obd_export *mdt_child_exp;
157 struct md_device *mdt_child;
158 struct dt_device *mdt_bottom;
159 struct obd_export *mdt_bottom_exp;
161 struct lu_target mdt_lut;
163 * Options bit-fields.
166 unsigned int mo_user_xattr:1,
174 /* mdt state flags */
175 unsigned long mdt_state;
176 /* lock to protect IOepoch */
177 spinlock_t mdt_ioepoch_lock;
180 /* transaction callbacks */
181 struct dt_txn_callback mdt_txn_cb;
183 /* these values should be updated from lov if necessary.
184 * or should be placed somewhere else. */
186 int mdt_max_cookiesize;
190 struct upcall_cache *mdt_identity_cache;
193 rwlock_t mdt_sptlrpc_lock;
194 struct sptlrpc_rule_set mdt_sptlrpc_rset;
196 /* capability keys */
197 unsigned long mdt_capa_timeout;
199 struct dt_object *mdt_ck_obj;
200 unsigned long mdt_ck_timeout;
201 unsigned long mdt_ck_expiry;
202 cfs_timer_t mdt_ck_timer;
203 struct ptlrpc_thread mdt_ck_thread;
204 struct lustre_capa_key mdt_capa_keys[2];
205 unsigned int mdt_capa_conf:1,
207 /* Enable remote dir on non-MDT0 */
208 mdt_enable_remote_dir:1;
210 gid_t mdt_enable_remote_dir_gid;
211 /* statfs optimization: we cache a bit */
212 struct obd_statfs mdt_osfs;
214 spinlock_t mdt_osfs_lock;
217 uid_t mdt_squash_uid;
218 gid_t mdt_squash_gid;
219 cfs_list_t mdt_nosquash_nids;
220 char *mdt_nosquash_str;
221 int mdt_nosquash_strlen;
222 struct rw_semaphore mdt_squash_sem;
225 struct rename_stats mdt_rename_stats;
226 struct lu_fid mdt_md_root_fid;
228 /* connection to quota master */
229 struct obd_export *mdt_qmt_exp;
230 /* quota master device associated with this MDT */
231 struct lu_device *mdt_qmt_dev;
233 struct coordinator mdt_coordinator;
236 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
237 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
238 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
239 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
240 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
241 #define MDT_COS_DEFAULT (0)
244 struct lu_object_header mot_header;
245 struct lu_object mot_obj;
248 int mot_ioepoch_count;
250 /* Lock to protect object's IO epoch. */
251 struct mutex mot_ioepoch_mutex;
252 /* Lock to protect create_data */
253 struct mutex mot_lov_mutex;
254 /* Lock to protect lease open.
255 * Lease open acquires write lock; normal open acquires read lock */
256 struct rw_semaphore mot_open_sem;
257 atomic_t mot_lease_count;
258 atomic_t mot_open_count;
259 /* A lock to protect EA data from racing setxattr and getxattrall */
260 struct rw_semaphore mot_xattr_sem;
263 enum mdt_object_flags {
264 /** SOM attributes are changed. */
265 MOF_SOM_CHANGE = (1 << 0),
267 * The SOM recovery state for mdt object.
268 * This state is an in-memory equivalent of an absent SOM EA, used
269 * instead of invalidating SOM EA while IOEpoch is still opened when
270 * a client eviction occurs or a client fails to obtain SOM attributes.
271 * It indicates that the last IOEpoch holder will need to obtain SOM
272 * attributes under [0;EOF] extent lock to flush all the client's
273 * cached of evicted from MDS clients (but not necessary evicted from
274 * OST) before taking ost attributes.
276 MOF_SOM_RECOV = (1 << 1),
277 /** File has been just created. */
278 MOF_SOM_CREATED = (1 << 2),
279 /** lov object has been created. */
280 MOF_LOV_CREATED = (1 << 3),
283 struct mdt_lock_handle {
284 /* Lock type, reg for cross-ref use or pdo lock. */
288 struct lustre_handle mlh_reg_lh;
289 ldlm_mode_t mlh_reg_mode;
292 struct lustre_handle mlh_pdo_lh;
293 ldlm_mode_t mlh_pdo_mode;
294 unsigned int mlh_pdo_hash;
296 /* Remote regular lock */
297 struct lustre_handle mlh_rreg_lh;
298 ldlm_mode_t mlh_rreg_mode;
302 MDT_LH_PARENT, /* parent lockh */
303 MDT_LH_CHILD, /* child lockh */
304 MDT_LH_OLD, /* old lockh for rename */
305 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
306 MDT_LH_NEW, /* new lockh for rename */
307 MDT_LH_RMT, /* used for return lh to caller */
308 MDT_LH_LOCAL, /* local lock never return to client */
317 struct mdt_reint_record {
318 mdt_reint_t rr_opcode;
319 const struct lustre_handle *rr_handle;
320 const struct lu_fid *rr_fid1;
321 const struct lu_fid *rr_fid2;
326 const void *rr_eadata;
329 const struct llog_cookie *rr_logcookies;
333 enum mdt_reint_flag {
334 MRF_OPEN_TRUNC = 1 << 0,
337 struct mdt_thread_info;
339 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
343 tx_exec_func_t exec_fn;
344 tx_exec_func_t undo_fn;
345 struct dt_object *object;
347 struct update_reply *reply;
352 const struct dt_rec *rec;
353 const struct dt_key *key;
368 struct dt_allocation_hint hint;
369 struct dt_object_format dof;
377 struct ost_body *body;
382 #define TX_MAX_OPS 10
383 struct thandle_exec_args {
384 struct thandle *ta_handle;
385 struct dt_device *ta_dev;
387 struct tx_arg ta_args[TX_MAX_OPS];
388 int ta_argno; /* used args */
392 * Common data shared by mdt-level handlers. This is allocated per-thread to
393 * reduce stack consumption.
395 struct mdt_thread_info {
398 * The following members will be filled explicitly
399 * with specific data in mdt_thread_info_init().
401 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
402 * request handling may migrate from one server thread to another.
404 struct req_capsule *mti_pill;
406 /* although we have export in req, there are cases when it is not
407 * available, e.g. closing files upon export destroy */
408 struct obd_export *mti_exp;
410 * A couple of lock handles.
412 struct mdt_lock_handle mti_lh[MDT_LH_NR];
414 struct mdt_device *mti_mdt;
415 const struct lu_env *mti_env;
418 * Additional fail id that can be set by handler. Passed to
419 * target_send_reply().
423 /* transaction number of current request */
429 * The following members will be filled expilictly
430 * with zero in mdt_thread_info_init(). These members may be used
437 struct md_attr mti_attr;
439 * Body for "habeo corpus" operations.
441 const struct mdt_body *mti_body;
443 * Host object. This is released at the end of mdt_handler().
445 struct mdt_object *mti_object;
447 * Lock request for "habeo clavis" operations.
449 const struct ldlm_request *mti_dlm_req;
451 __u32 mti_has_trans:1, /* has txn already? */
454 /* opdata for mdt_reint_open(), has the same as
455 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
456 * value onto disk for recovery when mdt_trans_stop_cb() is called.
462 * The following members will be filled explicitly
463 * with zero in mdt_reint_unpack(), because they are only used
464 * by reint requests (including mdt_reint_open()).
468 * reint record. contains information for reint operations.
470 struct mdt_reint_record mti_rr;
472 /** md objects included in operation */
473 struct mdt_object *mti_mos;
474 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
476 * Operation specification (currently create and lookup)
478 struct md_op_spec mti_spec;
482 * The following members will _NOT_ be initialized at all.
483 * DO NOT expect them to contain any valid value.
484 * They should be initialized explicitly by the user themselves.
487 /* XXX: If something is in a union, make sure they do not conflict */
489 struct lu_fid mti_tmp_fid1;
490 struct lu_fid mti_tmp_fid2;
491 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
492 * mdt_rename_lock() */
493 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
496 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
497 char ns_name[48]; /* for mdt_init0() */
498 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
499 struct obd_statfs osfs; /* for mdt_statfs() */
501 /* for mdt_readpage() */
502 struct lu_rdpg mti_rdpg;
503 /* for mdt_sendpage() */
504 struct l_wait_info mti_wait_info;
508 struct md_som_data data;
511 struct dt_object_format mti_update_dof;
512 struct update_reply *mti_update_reply;
513 struct update *mti_update;
514 int mti_update_reply_index;
515 struct obdo mti_obdo;
516 struct dt_object *mti_dt_object;
520 /* IO epoch related stuff. */
521 struct mdt_ioepoch *mti_ioepoch;
522 __u64 mti_replayepoch;
525 struct lu_buf mti_buf;
526 struct lu_buf mti_big_buf;
527 struct lustre_capa_key mti_capa_key;
529 /* Ops object filename */
530 struct lu_name mti_name;
531 /* per-thread values, can be re-used */
534 /* big_lmm buffer was used and must be used in reply */
535 int mti_big_lmm_used;
536 /* should be enough to fit lustre_mdt_attrs */
537 char mti_xattr_buf[128];
538 struct thandle_exec_args mti_handle;
539 struct ldlm_enqueue_info mti_einfo;
542 /* ptlrpc request handler for MDT. All handlers are
543 * grouped into several slices - struct mdt_opc_slice,
544 * and stored in an array - mdt_handlers[].
547 /* The name of this handler. */
549 /* Fail id for this handler, checked at the beginning of this handler*/
551 /* Operation code for this handler */
553 /* flags are listed in enum mdt_handler_flags below. */
555 /* The actual handler function to execute. */
556 int (*mh_act)(struct mdt_thread_info *info);
557 /* Request format for this request. */
558 const struct req_format *mh_fmt;
561 struct mdt_opc_slice {
564 struct mdt_handler *mos_hs;
567 struct cdt_req_progress {
568 struct mutex crp_lock; /**< protect tree */
569 struct interval_node *crp_root; /**< tree to track extent
571 struct interval_node **crp_node; /**< buffer for tree nodes
572 * vector of fixed size
574 int crp_cnt; /**< # of used nodes */
575 int crp_max; /**< # of allocated nodes */
578 struct cdt_agent_req {
579 cfs_list_t car_request_list; /**< to chain all the req. */
580 cfs_atomic_t car_refcount; /**< reference counter */
581 __u64 car_compound_id; /**< compound id */
582 __u64 car_flags; /**< request original flags */
583 struct obd_uuid car_uuid; /**< agent doing the req. */
584 __u32 car_archive_id; /**< archive id */
585 int car_canceled; /**< request was canceled */
586 cfs_time_t car_req_start; /**< start time */
587 cfs_time_t car_req_update; /**< last update time */
588 struct hsm_action_item *car_hai; /**< req. to the agent */
589 struct cdt_req_progress car_progress; /**< track data mvt
594 cfs_list_t ha_list; /**< to chain the agents */
595 struct obd_uuid ha_uuid; /**< agent uuid */
596 __u32 *ha_archive_id; /**< archive id */
597 int ha_archive_cnt; /**< number of archive entries
598 * 0 means any archive */
599 cfs_atomic_t ha_requests; /**< current request count */
600 cfs_atomic_t ha_success; /**< number of successful
602 cfs_atomic_t ha_failure; /**< number of failed actions */
605 struct cdt_restore_handle {
606 cfs_list_t crh_list; /**< to chain the handle */
607 struct lu_fid crh_fid; /**< fid of the object */
608 struct ldlm_extent crh_extent; /**< extent of the restore */
609 struct mdt_lock_handle crh_lh; /**< lock handle */
612 static inline const struct md_device_operations *
613 mdt_child_ops(struct mdt_device * m)
615 LASSERT(m->mdt_child);
616 return m->mdt_child->md_ops;
619 static inline struct md_object *mdt_object_child(struct mdt_object *o)
622 return lu2md(lu_object_next(&o->mot_obj));
625 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
627 return info->mti_pill ? info->mti_pill->rc_req : NULL;
630 static inline int req_is_replay(struct ptlrpc_request *req)
632 LASSERT(req->rq_reqmsg);
633 return !!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
636 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
638 LASSERT(info->mti_exp);
639 return exp_connect_flags(info->mti_exp);
642 static inline void mdt_object_get(const struct lu_env *env,
643 struct mdt_object *o)
646 lu_object_get(&o->mot_obj);
650 static inline void mdt_object_put(const struct lu_env *env,
651 struct mdt_object *o)
654 lu_object_put(env, &o->mot_obj);
658 static inline int mdt_object_exists(const struct mdt_object *o)
660 return lu_object_exists(&o->mot_obj);
663 static inline int mdt_object_remote(const struct mdt_object *o)
665 return lu_object_remote(&o->mot_obj);
668 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
670 return lu_object_fid(&o->mot_obj);
673 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
675 return mdt->mdt_lu_dev.ld_site;
678 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
680 return &mdt->mdt_seq_site;
683 static inline void mdt_export_evict(struct obd_export *exp)
685 class_fail_export(exp);
686 class_export_put(exp);
689 /* Here we use LVB_TYPE to check dne client, because it is
690 * also landed on 2.4. */
691 static inline int mdt_is_dne_client(struct obd_export *exp)
693 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
696 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
697 void mdt_set_disposition(struct mdt_thread_info *info,
698 struct ldlm_reply *rep, int flag);
699 void mdt_clear_disposition(struct mdt_thread_info *info,
700 struct ldlm_reply *rep, int flag);
702 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
703 ldlm_mode_t lm, const char *name,
706 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
709 int mdt_lock_setup(struct mdt_thread_info *info,
710 struct mdt_object *o,
711 struct mdt_lock_handle *lh);
713 int mdt_object_lock(struct mdt_thread_info *,
715 struct mdt_lock_handle *,
718 int mdt_object_lock_try(struct mdt_thread_info *,
720 struct mdt_lock_handle *,
723 void mdt_object_unlock(struct mdt_thread_info *,
725 struct mdt_lock_handle *,
728 struct mdt_object *mdt_object_new(const struct lu_env *,
730 const struct lu_fid *);
731 struct mdt_object *mdt_object_find(const struct lu_env *,
733 const struct lu_fid *);
734 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
735 const struct lu_fid *,
736 struct mdt_lock_handle *,
738 void mdt_object_unlock_put(struct mdt_thread_info *,
740 struct mdt_lock_handle *,
743 void mdt_client_compatibility(struct mdt_thread_info *info);
745 int mdt_remote_object_lock(struct mdt_thread_info *mti,
746 struct mdt_object *o, struct lustre_handle *lh,
747 ldlm_mode_t mode, __u64 ibits);
748 int mdt_close_unpack(struct mdt_thread_info *info);
749 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
750 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
751 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
752 const struct lu_attr *attr, const struct lu_fid *fid);
754 int mdt_getxattr(struct mdt_thread_info *info);
755 int mdt_reint_setxattr(struct mdt_thread_info *info,
756 struct mdt_lock_handle *lh);
758 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
759 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
761 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
762 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
763 struct mdt_lock_handle *lhc);
765 extern void target_recovery_fini(struct obd_device *obd);
766 extern void target_recovery_init(struct lu_target *lut,
767 svc_handler_t handler);
768 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
769 struct obd_device *, struct lustre_sb_info *lsi);
770 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
772 int mdt_export_stats_init(struct obd_device *obd,
773 struct obd_export *exp,
776 int mdt_pin(struct mdt_thread_info* info);
778 int mdt_lock_new_child(struct mdt_thread_info *info,
779 struct mdt_object *o,
780 struct mdt_lock_handle *child_lockh);
782 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
785 int mdt_reint_open(struct mdt_thread_info *info,
786 struct mdt_lock_handle *lhc);
788 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
789 const struct lustre_handle *handle,
793 MDT_IOEPOCH_CLOSED = 0,
794 MDT_IOEPOCH_OPENED = 1,
795 MDT_IOEPOCH_GETATTR = 2,
803 int mdt_attr_get_complex(struct mdt_thread_info *info,
804 struct mdt_object *o, struct md_attr *ma);
805 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
807 int mdt_object_is_som_enabled(struct mdt_object *mo);
808 int mdt_write_get(struct mdt_object *o);
809 void mdt_write_put(struct mdt_object *o);
810 int mdt_write_read(struct mdt_object *o);
811 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
812 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
813 void mdt_mfd_free(struct mdt_file_data *mfd);
814 int mdt_close(struct mdt_thread_info *info);
815 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
816 struct md_attr *ma, int flags);
817 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
819 int mdt_done_writing(struct mdt_thread_info *info);
820 int mdt_fix_reply(struct mdt_thread_info *info);
821 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
822 const struct md_attr *);
823 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
825 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
826 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
827 const void *area, ssize_t len);
829 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
831 int mdt_check_ucred(struct mdt_thread_info *);
832 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
833 int mdt_init_ucred_reint(struct mdt_thread_info *);
834 void mdt_exit_ucred(struct mdt_thread_info *);
835 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
836 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
837 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
839 int mdt_handle_common(struct ptlrpc_request *req,
840 struct mdt_opc_slice *supported);
841 int mdt_connect(struct mdt_thread_info *info);
842 int mdt_disconnect(struct mdt_thread_info *info);
843 int mdt_set_info(struct mdt_thread_info *info);
844 int mdt_get_info(struct mdt_thread_info *info);
845 int mdt_getstatus(struct mdt_thread_info *info);
846 int mdt_getattr(struct mdt_thread_info *info);
847 int mdt_getattr_name(struct mdt_thread_info *info);
848 int mdt_statfs(struct mdt_thread_info *info);
849 int mdt_reint(struct mdt_thread_info *info);
850 int mdt_sync(struct mdt_thread_info *info);
851 int mdt_is_subdir(struct mdt_thread_info *info);
852 int mdt_obd_ping(struct mdt_thread_info *info);
853 int mdt_obd_log_cancel(struct mdt_thread_info *info);
854 int mdt_obd_qc_callback(struct mdt_thread_info *info);
855 int mdt_enqueue(struct mdt_thread_info *info);
856 int mdt_convert(struct mdt_thread_info *info);
857 int mdt_bl_callback(struct mdt_thread_info *info);
858 int mdt_cp_callback(struct mdt_thread_info *info);
859 int mdt_llog_create(struct mdt_thread_info *info);
860 int mdt_llog_destroy(struct mdt_thread_info *info);
861 int mdt_llog_read_header(struct mdt_thread_info *info);
862 int mdt_llog_next_block(struct mdt_thread_info *info);
863 int mdt_llog_prev_block(struct mdt_thread_info *info);
864 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
865 int mdt_readpage(struct mdt_thread_info *info);
866 int mdt_obd_idx_read(struct mdt_thread_info *info);
868 extern struct mdt_opc_slice mdt_regular_handlers[];
869 extern struct mdt_opc_slice mdt_seq_handlers[];
870 extern struct mdt_opc_slice mdt_fld_handlers[];
872 int mdt_quotacheck(struct mdt_thread_info *info);
873 int mdt_quotactl(struct mdt_thread_info *info);
874 int mdt_quota_dqacq(struct mdt_thread_info *info);
875 int mdt_swap_layouts(struct mdt_thread_info *info);
877 extern struct lprocfs_vars lprocfs_mds_module_vars[];
878 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
880 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
883 struct mdt_handler *mdt_handler_find(__u32 opc,
884 struct mdt_opc_slice *supported);
886 int mdt_init_sec_level(struct mdt_thread_info *);
887 int mdt_init_idmap(struct mdt_thread_info *);
888 void mdt_cleanup_idmap(struct mdt_export_data *);
889 int mdt_handle_idmap(struct mdt_thread_info *);
890 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
891 struct ptlrpc_user_desc *);
892 void mdt_body_reverse_idmap(struct mdt_thread_info *,
894 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
895 struct mdt_remote_perm *);
896 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
898 static inline struct mdt_device *mdt_dev(struct lu_device *d)
900 return container_of0(d, struct mdt_device, mdt_lu_dev);
903 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
905 struct lu_object *lo;
906 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
908 lo = lu_object_locate(mo->mot_obj.lo_header,
909 mdt->mdt_bottom->dd_lu_dev.ld_type);
914 /* mdt/mdt_identity.c */
915 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
917 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
919 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
921 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
923 void mdt_flush_identity(struct upcall_cache *, int);
925 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
927 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
930 int mdt_hsm_state_get(struct mdt_thread_info *info);
931 int mdt_hsm_state_set(struct mdt_thread_info *info);
932 int mdt_hsm_action(struct mdt_thread_info *info);
933 int mdt_hsm_progress(struct mdt_thread_info *info);
934 int mdt_hsm_ct_register(struct mdt_thread_info *info);
935 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
936 int mdt_hsm_request(struct mdt_thread_info *info);
937 /* mdt/mdt_hsm_cdt_actions.c */
938 extern const struct file_operations mdt_agent_actions_fops;
939 void dump_llog_agent_req_rec(char *prefix, struct llog_agent_req_rec *larr);
940 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
941 llog_cb_t cb, void *data);
942 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
943 __u64 compound_id, __u32 archive_id,
944 __u64 flags, struct hsm_action_item *hai);
945 int mdt_agent_record_update(const struct lu_env *env,
946 struct mdt_device *mdt, __u64 *cookies,
947 int cookies_count, enum agent_req_status status);
948 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
949 struct llog_handle *llh,
950 struct llog_agent_req_rec *larr);
952 /* mdt/mdt_hsm_cdt_agent.c */
953 extern const struct file_operations mdt_hsm_agent_fops;
954 int mdt_hsm_agent_register(struct mdt_thread_info *info,
955 const struct obd_uuid *uuid,
956 int nr_archives, __u32 *archive_num);
957 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
958 const struct obd_uuid *uuid,
960 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
961 const struct obd_uuid *uuid);
962 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
963 int succ_rq, int fail_rq, int new_rq,
964 const struct obd_uuid *uuid);
965 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
966 struct obd_uuid *uuid);
967 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
969 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
970 struct hsm_progress_kernel *pgs);
971 /* mdt/mdt_hsm_cdt_client.c */
972 int mdt_hsm_add_actions(struct mdt_thread_info *info,
973 struct hsm_action_list *hal, __u64 *compound_id);
974 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
975 struct hsm_action_list *hal);
976 int mdt_hsm_get_running(struct mdt_thread_info *mti,
977 struct hsm_action_list *hal);
978 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
979 const struct lu_fid *fid);
981 /* mdt/mdt_hsm_cdt_requests.c */
982 extern const struct file_operations mdt_hsm_request_fops;
983 void dump_requests(char *prefix, struct coordinator *cdt);
984 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
985 __u64 flags, struct obd_uuid *uuid,
986 struct hsm_action_item *hai);
987 void mdt_cdt_free_request(struct cdt_agent_req *car);
988 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
989 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
991 const struct lu_fid *fid);
992 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
993 void mdt_cdt_get_request(struct cdt_agent_req *car);
994 void mdt_cdt_put_request(struct cdt_agent_req *car);
995 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
996 struct hsm_progress_kernel *pgs);
997 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
999 /* fake functions, will be remove with patch LU-3343 */
1000 static inline struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1003 struct mdt_lock_handle *lh)
1005 return ERR_PTR(-EINVAL);
1007 static inline bool mdt_hsm_is_action_compat(struct hsm_action_item *hai,
1008 int hal_an, __u64 rq_flags,
1013 static inline int mdt_hsm_cdt_init(struct mdt_device *mdt)
1015 struct coordinator *cdt = &mdt->mdt_coordinator;
1017 /* minimal init before final patch landing */
1018 sema_init(&cdt->cdt_llog_lock, 1);
1019 init_rwsem(&cdt->cdt_agent_lock);
1020 init_rwsem(&cdt->cdt_request_lock);
1021 sema_init(&cdt->cdt_restore_lock, 1);
1023 CFS_INIT_LIST_HEAD(&cdt->cdt_requests);
1024 CFS_INIT_LIST_HEAD(&cdt->cdt_agents);
1025 CFS_INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
1027 cdt->cdt_state = CDT_STOPPED;
1030 static inline int mdt_hsm_cdt_start(struct mdt_device *mdt)
1034 static inline int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1038 static inline int mdt_hsm_cdt_fini(struct mdt_device *mdt)
1042 static inline int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
1046 static inline int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1047 struct hsm_action_list *hal,
1048 const struct obd_uuid *uuid)
1052 static inline int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1053 struct hsm_progress_kernel *pgs,
1058 /* end of fake functions */
1060 extern struct lu_context_key mdt_thread_key;
1061 /* debug issues helper starts here*/
1062 static inline int mdt_fail_write(const struct lu_env *env,
1063 struct dt_device *dd, int id)
1065 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1066 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1068 return dd->dd_ops->dt_ro(env, dd);
1069 /* We set FAIL_ONCE because we never "un-fail" a device */
1075 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1077 return &req->rq_export->exp_mdt_data;
1080 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1081 struct mdt_lock_handle *lhc);
1082 static inline int mdt_check_resent(struct mdt_thread_info *info,
1083 mdt_reconstruct_t reconstruct,
1084 struct mdt_lock_handle *lhc)
1086 struct ptlrpc_request *req = mdt_info_req(info);
1089 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1090 if (req_xid_is_last(req)) {
1091 reconstruct(info, lhc);
1094 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
1095 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
1100 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1101 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1103 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1105 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1108 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1110 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1111 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1112 struct lustre_handle *lh,
1114 ldlm_policy_data_t *policy,
1115 const struct ldlm_res_id *res_id,
1116 __u64 flags, const __u64 *client_cookie)
1120 LASSERT(ns != NULL);
1121 LASSERT(lh != NULL);
1123 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1124 mode, &flags, mdt_blocking_ast,
1125 ldlm_completion_ast, NULL, NULL, 0,
1126 LVB_T_NONE, client_cookie, lh);
1127 return rc == ELDLM_OK ? 0 : -EIO;
1130 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1133 ldlm_lock_decref(lh, mode);
1136 extern mdl_mode_t mdt_mdl_lock_modes[];
1137 extern ldlm_mode_t mdt_dlm_lock_modes[];
1139 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1141 LASSERT(IS_PO2(mode));
1142 return mdt_mdl_lock_modes[mode];
1145 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1147 LASSERT(IS_PO2(mode));
1148 return mdt_dlm_lock_modes[mode];
1152 extern struct ldlm_valblock_ops mdt_lvbo;
1154 static inline struct lu_name *mdt_name(const struct lu_env *env,
1155 char *name, int namelen)
1157 struct lu_name *lname;
1158 struct mdt_thread_info *mti;
1160 LASSERT(namelen > 0);
1161 /* trailing '\0' in buffer */
1162 LASSERT(name[namelen] == '\0');
1164 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
1165 lname = &mti->mti_name;
1166 lname->ln_name = name;
1167 lname->ln_namelen = namelen;
1171 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
1172 struct lu_name *slname)
1177 tlname->ln_name = slname->ln_name;
1178 tlname->ln_namelen = slname->ln_namelen;
1182 void mdt_enable_cos(struct mdt_device *, int);
1183 int mdt_cos_is_enabled(struct mdt_device *);
1201 LPROC_MDT_SAMEDIR_RENAME,
1202 LPROC_MDT_CROSSDIR_RENAME,
1205 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1206 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1207 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1208 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1209 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1210 void mdt_procfs_fini(struct mdt_device *mdt);
1211 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1212 struct mdt_device *mdt,
1213 struct ptlrpc_request *req,
1214 struct mdt_object *src, struct mdt_object *tgt);
1217 int mdt_ck_thread_start(struct mdt_device *mdt);
1218 void mdt_ck_thread_stop(struct mdt_device *mdt);
1219 void mdt_ck_timer_callback(unsigned long castmeharder);
1220 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1222 static inline void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1223 const struct lu_fid *fid,
1224 struct lustre_capa *capa)
1226 struct md_capainfo *ci;
1228 LASSERT(offset >= 0 && offset < MD_CAPAINFO_MAX);
1229 if (!info->mti_mdt->mdt_opts.mo_mds_capa ||
1230 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
1233 ci = md_capainfo(info->mti_env);
1235 ci->mc_fid[offset] = *fid;
1236 ci->mc_capa[offset] = capa;
1239 static inline void mdt_dump_capainfo(struct mdt_thread_info *info)
1241 struct md_capainfo *ci = md_capainfo(info->mti_env);
1246 for (i = 0; i < MD_CAPAINFO_MAX; i++) {
1247 if (!ci->mc_capa[i]) {
1248 CERROR("no capa for index %d "DFID"\n",
1249 i, PFID(&ci->mc_fid[i]));
1252 if (ci->mc_capa[i] == BYPASS_CAPA) {
1253 CERROR("bypass for index %d "DFID"\n",
1254 i, PFID(&ci->mc_fid[i]));
1257 DEBUG_CAPA(D_ERROR, ci->mc_capa[i], "index %d", i);
1261 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1263 return mdt->mdt_lu_dev.ld_obd;
1266 extern const struct lu_device_operations mdt_lu_ops;
1268 static inline int lu_device_is_mdt(struct lu_device *d)
1270 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1273 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1275 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1276 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1277 return container_of0(d, struct mdt_device, mdt_lu_dev);
1280 static inline char *mdt_obd_name(struct mdt_device *mdt)
1282 return mdt->mdt_lu_dev.ld_obd->obd_name;
1285 int mds_mod_init(void);
1286 void mds_mod_exit(void);
1288 /* Update handlers */
1289 int out_handle(struct mdt_thread_info *info);
1291 #define out_tx_create(info, obj, attr, fid, dof, th, reply, idx) \
1292 __out_tx_create(info, obj, attr, fid, dof, th, reply, idx, \
1295 #define out_tx_attr_set(info, obj, attr, th, reply, idx) \
1296 __out_tx_attr_set(info, obj, attr, th, reply, idx, \
1299 #define out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx) \
1300 __out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx, \
1303 #define out_tx_ref_add(info, obj, th, reply, idx) \
1304 __out_tx_ref_add(info, obj, th, reply, idx, __FILE__, __LINE__)
1306 #define out_tx_ref_del(info, obj, th, reply, idx) \
1307 __out_tx_ref_del(info, obj, th, reply, idx, __FILE__, __LINE__)
1309 #define out_tx_index_insert(info, obj, th, name, fid, reply, idx) \
1310 __out_tx_index_insert(info, obj, th, name, fid, reply, idx, \
1313 #define out_tx_index_delete(info, obj, th, name, reply, idx) \
1314 __out_tx_index_delete(info, obj, th, name, reply, idx, \
1317 #define out_tx_destroy(info, obj, th, reply, idx) \
1318 __out_tx_destroy(info, obj, th, reply, idx, __FILE__, __LINE__)
1320 #endif /* __KERNEL__ */